BlobGAN: Spatially Disentangled Scene Representations

[joyhuang9473]

BlobGAN: Spatially Disentangled Scene Representations

• Author: Dave Epstein, Taesung Park, Richard Zhang, Eli Shechtman, Alexei A. Efros
• Origin: https://arxiv.org/abs/2205.02837
• Related:
  • project page
  • github
  • video

We propose an unsupervised, mid-level representation for a generative model of scenes.

our network learns to associate different blobs with different entities in a scene and to arrange these blobs to capture scene layout.